Valuation Methodology
Overview
AI Investor Barometer uses a deterministic valuation engine to convert AI-generated assumptions into model estimates. Each AI model produces 4 core assumptions; the engine applies a standardized calculation to produce a comparable estimate.
Two model types are used depending on company sector:
- A. Two-Stage FCFF DCF — for all non-financial sectors
- B. Excess Return on Equity — for banks and insurance
AI Model Inputs
Each AI model independently produces these assumptions after analyzing public financial data:
| Parameter | Description | Example |
|---|---|---|
| revenue_cagr_5y | 5-year revenue growth rate | 6% |
| ebit_margin_target | Long-term EBIT margin target | 14% |
| wacc | Weighted average cost of capital | 8.5% |
| terminal_growth | Perpetuity growth rate | 2% |
Model 1: Two-Stage FCFF DCF
Used for technology, industrials, energy, healthcare, consumer, materials, telecom, utilities, and real estate sectors. The model projects free cash flows in three phases.
Phase 1 — Explicit Period
8–10 years of explicit revenue and margin projection. Revenue grows at the AI model's CAGR assumption. EBIT margin ramps concavely from current TTM margin to target margin.
Phase 2 — Fade Period
3 years where revenue growth fades linearly from CAGR to terminal growth rate. Margin stays at target.
Phase 3 — Terminal Value
Gordon Growth Model on the last fade-year FCF.
Margin Ramp — Concave Interpolation
The margin ramps from current to target using a concave function, where higher concavity values mean faster initial ramp:
Free Cash Flow Calculation
ROIC-based reinvestment: higher growth requires more reinvestment, reducing FCF conversion.
Uses actual CapEx and D&A rates from financial statements. CapEx normalizes toward sector typical during fade period.
Terminal Value
Gordon Growth Model with minimum WACC−g spread of 3.5pp:
Enterprise Value → Estimate
Sum discounted FCFs + discounted terminal value, subtract net debt, divide by shares outstanding.
Model 2: Excess Return on Equity
Used for financial sector companies (banks, insurance). Net debt is NOT subtracted — liabilities are the business.
ROE ramps concavely from current to target over the explicit period. In the fade period, ROE converges halfway toward cost of equity.
Excess return each year = Net Income − Equity × Cost of Equity. Retained earnings grow book equity.
Insurance adjustment: combined ratio > 100% reduces achievable ROE via underwriting drag.
Terminal excess return is capitalized as a perpetuity at cost of equity.
Validation & Clipping
All AI model assumptions are clipped to sector-specific bounds before computation. This prevents extreme or nonsensical inputs from producing meaningless estimates.
CAPM WACC Anchoring
When company beta is available (0.3 ≤ β ≤ 3.0), WACC is anchored to ±2pp of a CAPM-derived estimate:
Safety Caps (Post-Valuation)
Four layers of caps prevent extreme estimates:
Bayesian Calibration
After all safety caps, the engine blends the DCF estimate with analyst consensus using a fixed shrinkage weight. This is a cold-start calibration measure to reduce systematic bias while the platform accumulates its own backtest history.
The current weight α = 0.70 means 70% of the final estimate comes from the DCF model and 30% from analyst consensus. This weight is temporary and will be empirically optimized per model once 6–12 months of data is available.
Importantly, the AI Market Indices (ACDI, ADI) are computed from the pre-calibration DCF estimates to preserve the pure model signal. The calibrated estimate is a presentation-layer feature, not the analytical foundation.
Sector Profiles
Each sector has calibrated bounds for all assumptions. Values outside these ranges are clipped.
| Sector | CAGR | Margin | WACC | g | FCF | CpEx | D&A | Conc. | Yr |
|---|---|---|---|---|---|---|---|---|---|
| Technology | −10 / +30% | −10 / +55% | 7 / 14% | 3% | 0.80 | 3% | 3% | 1.8 | 8+3 |
| Telecom | −2 / +6% | 5 / 35% | 5 / 10% | 2% | 0.45 | 16% | 13% | 1.1 | 10+3 |
| Industrials | −10 / +15% | 2 / 20% | 6 / 12% | 2.5% | 0.60 | 6% | 5% | 1.3 | 8+3 |
| Materials | −10 / +10% | 0 / 22% | 7 / 13% | 2% | 0.50 | 9% | 7% | 1.2 | 8+3 |
| Energy | −5 / +12% | 2 / 22% | 7 / 13% | 2% | 0.50 | 10% | 6% | 1.2 | 8+3 |
| Consumer | −5 / +15% | 2 / 25% | 6 / 12% | 2.5% | 0.70 | 4% | 4% | 1.4 | 8+3 |
| Healthcare | −3 / +12% | 5 / +45% | 7 / 12% | 2.5% | 0.75 | 4% | 5% | 1.3 | 7+3 |
| Utilities | −2 / +8% | 5 / 30% | 4 / 10% | 2% | 0.50 | 12% | 8% | 1.1 | 10+3 |
| Real Estate | −5 / +10% | 10 / 30% | 5 / 12% | 2% | 0.55 | 8% | 10% | 1.1 | 8+3 |
| Financials | −5 / +10% | 10 / 50% | 6 / 16% | 2.5% | — | — | — | — | 8+2 |
P/E Cap Ranges
| Sector | P/E Low | P/E High | Cap (1.5×) |
|---|---|---|---|
| Technology | 18× | 35× | 52.5× |
| Telecom | 12× | 20× | 30× |
| Industrials | 15× | 25× | 37.5× |
| Materials | 10× | 18× | 27× |
| Energy | 10× | 18× | 27× |
| Consumer | 15× | 25× | 37.5× |
| Healthcare | 18× | 28× | 42× |
| Utilities | 14× | 22× | 33× |
| Financials | 10× | 16× | 24× |
TV Multiple Caps
| Sector | TV Cap |
|---|---|
| Technology | 30× |
| Telecom | 20× |
| Industrials | 22× |
| Materials | 18× |
| Energy | 18× |
| Consumer | 22× |
| Healthcare | 25× |
| Utilities | 20× |
| Real Estate | 22× |
Valuation Flags
Each estimate includes diagnostic flags indicating which caps and adjustments were applied:
| Flag | Description |
|---|---|
| clipped_fields | Which LLM assumptions were clipped to sector bounds |
| capm_wacc_anchor | CAPM-derived WACC anchor was applied |
| forced_g_below_wacc | Terminal growth forced down to maintain WACC−g ≥ 3.5pp |
| fcf_mode | FCF calculation mode: proxy_real or simplified |
| fcf_conv_roic_applied | ROIC-based FCF conversion was tighter than default |
| tv_capped | Terminal value hit sector multiple cap |
| tv_share_capped | Terminal value exceeded 88% of EV — capped |
| pe_capped | Forward P/E hard cap applied |
| analyst_tp_capped | Estimate clipped to ±40% of analyst consensus |
| capex_normalization | CapEx rate normalized toward sector typical during fade |
AI Market Indices
The platform computes three composite indices from model outputs — no additional LLM calls required. All indices are computed daily from existing pipeline data.
Company Universe
The barometer covers a curated universe of 24 companies across two markets. The universe is defined in a central configuration file and reviewed quarterly.
Finland (OMXH) — 12 companies
12 Finnish OMXH companies — largest by market cap with sufficient analyst coverage (≥5 analysts), public financials, and available spot price data. Subset of the OMXH25 index.
United States — 12 companies
12 US mega-cap companies representing major sectors. Selected as a benchmark to validate model behavior across different markets and currencies.
Addition Rules
New companies require: analyst coverage ≥5, public financials on Yahoo Finance, spot price from at least one source (Yahoo/Finnhub/Stooq), and a matching sector profile in the valuation engine.
Removal Rules
Companies are deactivated (never deleted) to preserve historical data. Reasons include: delisting, merger, insufficient data quality, or universe rebalancing.
Data Retention & Reproducibility
All model outputs, valuation results, and consensus data are retained indefinitely with no automated deletion. For every company on each trading day, the system stores: raw LLM assumptions (before clipping), clipped assumptions used in valuation, full prompt text and raw model responses, spot prices with source attribution, all diagnostic flags and valuation metadata (engine version, sector profile hash), and pipeline execution metadata (cost, latency, token counts). Universe changes (company activations/deactivations) are logged with timestamps and reasons. This complete audit trail enables full reproducibility of any historical valuation.